net/bridge/br_input.c

   1 /*
   2  *      Handle incoming frames
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/kernel.h>
  16 #include <linux/netdevice.h>
  17 #include <linux/etherdevice.h>
  18 #include <linux/netfilter_bridge.h>
  19 #include <linux/neighbour.h>
  20 #include <net/arp.h>
  21 #include <linux/export.h>
  22 #include <linux/rculist.h>
  23 #include "br_private.h"
  24 #include "br_private_tunnel.h"
  25
  26 /* Hook for brouter */
  27 br_should_route_hook_t __rcu *br_should_route_hook __read_mostly;
  28 EXPORT_SYMBOL(br_should_route_hook);
  29
  30 static int
  31 br_netif_receive_skb(struct net *net, struct sock *sk, struct sk_buff *skb)
  32 {
  33         br_drop_fake_rtable(skb);
  34         return netif_receive_skb(skb);
  35 }
  36
  37 static int br_pass_frame_up(struct sk_buff *skb)
  38 {
  39         struct net_device *indev, *brdev = BR_INPUT_SKB_CB(skb)->brdev;
  40         struct net_bridge *br = netdev_priv(brdev);
  41         struct net_bridge_vlan_group *vg;
  42         struct pcpu_sw_netstats *brstats = this_cpu_ptr(br->stats);
  43
  44         u64_stats_update_begin(&brstats->syncp);
  45         brstats->rx_packets++;
  46         brstats->rx_bytes += skb->len;
  47         u64_stats_update_end(&brstats->syncp);
  48
  49         vg = br_vlan_group_rcu(br);
  50
  51         /* Reset the offload_fwd_mark because there could be a stacked
  52          * bridge above, and it should not think this bridge it doing
  53          * that bridge's work forwarding out its ports.
  54          */
  55         br_switchdev_frame_unmark(skb);
  56
  57         /* Bridge is just like any other port.  Make sure the
  58          * packet is allowed except in promisc modue when someone
  59          * may be running packet capture.
  60          */
  61         if (!(brdev->flags & IFF_PROMISC) &&
  62             !br_allowed_egress(vg, skb)) {
  63                 kfree_skb(skb);
  64                 return NET_RX_DROP;
  65         }
  66
  67         indev = skb->dev;
  68         skb->dev = brdev;
  69         skb = br_handle_vlan(br, NULL, vg, skb);
  70         if (!skb)
  71                 return NET_RX_DROP;
  72         /* update the multicast stats if the packet is IGMP/MLD */
  73         br_multicast_count(br, NULL, skb, br_multicast_igmp_type(skb),
  74                            BR_MCAST_DIR_TX);
  75
  76         return NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
  77                        dev_net(indev), NULL, skb, indev, NULL,
  78                        br_netif_receive_skb);
  79 }
  80
  81 /* note: already called with rcu_read_lock */
  82 int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  83 {
  84         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
  85         enum br_pkt_type pkt_type = BR_PKT_UNICAST;
  86         struct net_bridge_fdb_entry *dst = NULL;
  87         struct net_bridge_mdb_entry *mdst;
  88         bool local_rcv, mcast_hit = false;
  89         struct net_bridge *br;
  90         u16 vid = 0;
  91
  92         if (!p || p->state == BR_STATE_DISABLED)
  93                 goto drop;
  94
  95         if (!br_allowed_ingress(p->br, nbp_vlan_group_rcu(p), skb, &vid))
  96                 goto out;
  97
  98         nbp_switchdev_frame_mark(p, skb);
  99
 100         /* insert into forwarding database after filtering to avoid spoofing */
 101         br = p->br;
 102         if (p->flags & BR_LEARNING)
 103                 br_fdb_update(br, p, eth_hdr(skb)->h_source, vid, false);
 104
 105         local_rcv = !!(br->dev->flags & IFF_PROMISC);
 106         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) {
 107                 /* by definition the broadcast is also a multicast address */
 108                 if (is_broadcast_ether_addr(eth_hdr(skb)->h_dest)) {
 109                         pkt_type = BR_PKT_BROADCAST;
 110                         local_rcv = true;
 111                 } else {
 112                         pkt_type = BR_PKT_MULTICAST;
 113                         if (br_multicast_rcv(br, p, skb, vid))
 114                                 goto drop;
 115                 }
 116         }
 117
 118         if (p->state == BR_STATE_LEARNING)
 119                 goto drop;
 120
 121         BR_INPUT_SKB_CB(skb)->brdev = br->dev;
 122         BR_INPUT_SKB_CB(skb)->src_port_isolated = !!(p->flags & BR_ISOLATED);
 123
 124         if (IS_ENABLED(CONFIG_INET) &&
 125             (skb->protocol == htons(ETH_P_ARP) ||
 126              skb->protocol == htons(ETH_P_RARP))) {
 127                 br_do_proxy_suppress_arp(skb, br, vid, p);
 128         } else if (IS_ENABLED(CONFIG_IPV6) &&
 129                    skb->protocol == htons(ETH_P_IPV6) &&
 130                    br->neigh_suppress_enabled &&
 131                    pskb_may_pull(skb, sizeof(struct ipv6hdr) +
 132                                  sizeof(struct nd_msg)) &&
 133                    ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
 134                         struct nd_msg *msg, _msg;
 135
 136                         msg = br_is_nd_neigh_msg(skb, &_msg);
 137                         if (msg)
 138                                 br_do_suppress_nd(skb, br, vid, p, msg);
 139         }
 140
 141         switch (pkt_type) {
 142         case BR_PKT_MULTICAST:
 143                 mdst = br_mdb_get(br, skb, vid);
 144                 if ((mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) &&
 145                     br_multicast_querier_exists(br, eth_hdr(skb))) {
 146                         if ((mdst && mdst->host_joined) ||
 147                             br_multicast_is_router(br)) {
 148                                 local_rcv = true;
 149                                 br->dev->stats.multicast++;
 150                         }
 151                         mcast_hit = true;
 152                 } else {
 153                         local_rcv = true;
 154                         br->dev->stats.multicast++;
 155                 }
 156                 break;
 157         case BR_PKT_UNICAST:
 158                 dst = br_fdb_find_rcu(br, eth_hdr(skb)->h_dest, vid);
 159         default:
 160                 break;
 161         }
 162
 163         if (dst) {
 164                 unsigned long now = jiffies;
 165
 166                 if (dst->is_local)
 167                         return br_pass_frame_up(skb);
 168
 169                 if (now != dst->used)
 170                         dst->used = now;
 171                 br_forward(dst->dst, skb, local_rcv, false);
 172         } else {
 173                 if (!mcast_hit)
 174                         br_flood(br, skb, pkt_type, local_rcv, false);
 175                 else
 176                         br_multicast_flood(mdst, skb, local_rcv, false);
 177         }
 178
 179         if (local_rcv)
 180                 return br_pass_frame_up(skb);
 181
 182 out:
 183         return 0;
 184 drop:
 185         kfree_skb(skb);
 186         goto out;
 187 }
 188 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
 189
 190 static void __br_handle_local_finish(struct sk_buff *skb)
 191 {
 192         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
 193         u16 vid = 0;
 194
 195         /* check if vlan is allowed, to avoid spoofing */
 196         if (p->flags & BR_LEARNING && br_should_learn(p, skb, &vid))
 197                 br_fdb_update(p->br, p, eth_hdr(skb)->h_source, vid, false);
 198 }
 199
 200 /* note: already called with rcu_read_lock */
 201 static int br_handle_local_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 202 {
 203         __br_handle_local_finish(skb);
 204
 205         /* return 1 to signal the okfn() was called so it's ok to use the skb */
 206         return 1;
 207 }
 208
 209 /*
 210  * Return NULL if skb is handled
 211  * note: already called with rcu_read_lock
 212  */
 213 rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
 214 {
 215         struct net_bridge_port *p;
 216         struct sk_buff *skb = *pskb;
 217         const unsigned char *dest = eth_hdr(skb)->h_dest;
 218         br_should_route_hook_t *rhook;
 219
 220         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 221                 return RX_HANDLER_PASS;
 222
 223         if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
 224                 goto drop;
 225
 226         skb = skb_share_check(skb, GFP_ATOMIC);
 227         if (!skb)
 228                 return RX_HANDLER_CONSUMED;
 229
 230         p = br_port_get_rcu(skb->dev);
 231         if (p->flags & BR_VLAN_TUNNEL) {
 232                 if (br_handle_ingress_vlan_tunnel(skb, p,
 233                                                   nbp_vlan_group_rcu(p)))
 234                         goto drop;
 235         }
 236
 237         if (unlikely(is_link_local_ether_addr(dest))) {
 238                 u16 fwd_mask = p->br->group_fwd_mask_required;
 239
 240                 /*
 241                  * See IEEE 802.1D Table 7-10 Reserved addresses
 242                  *
 243                  * Assignment                           Value
 244                  * Bridge Group Address         01-80-C2-00-00-00
 245                  * (MAC Control) 802.3          01-80-C2-00-00-01
 246                  * (Link Aggregation) 802.3     01-80-C2-00-00-02
 247                  * 802.1X PAE address           01-80-C2-00-00-03
 248                  *
 249                  * 802.1AB LLDP                 01-80-C2-00-00-0E
 250                  *
 251                  * Others reserved for future standardization
 252                  */
 253                 fwd_mask |= p->group_fwd_mask;
 254                 switch (dest[5]) {
 255                 case 0x00:      /* Bridge Group Address */
 256                         /* If STP is turned off,
 257                            then must forward to keep loop detection */
 258                         if (p->br->stp_enabled == BR_NO_STP ||
 259                             fwd_mask & (1u << dest[5]))
 260                                 goto forward;
 261                         *pskb = skb;
 262                         __br_handle_local_finish(skb);
 263                         return RX_HANDLER_PASS;
 264
 265                 case 0x01:      /* IEEE MAC (Pause) */
 266                         goto drop;
 267
 268                 case 0x0E:      /* 802.1AB LLDP */
 269                         fwd_mask |= p->br->group_fwd_mask;
 270                         if (fwd_mask & (1u << dest[5]))
 271                                 goto forward;
 272                         *pskb = skb;
 273                         __br_handle_local_finish(skb);
 274                         return RX_HANDLER_PASS;
 275
 276                 default:
 277                         /* Allow selective forwarding for most other protocols */
 278                         fwd_mask |= p->br->group_fwd_mask;
 279                         if (fwd_mask & (1u << dest[5]))
 280                                 goto forward;
 281                 }
 282
 283                 /* The else clause should be hit when nf_hook():
 284                  *   - returns < 0 (drop/error)
 285                  *   - returns = 0 (stolen/nf_queue)
 286                  * Thus return 1 from the okfn() to signal the skb is ok to pass
 287                  */
 288                 if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
 289                             dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 290                             br_handle_local_finish) == 1) {
 291                         return RX_HANDLER_PASS;
 292                 } else {
 293                         return RX_HANDLER_CONSUMED;
 294                 }
 295         }
 296
 297 forward:
 298         switch (p->state) {
 299         case BR_STATE_FORWARDING:
 300                 rhook = rcu_dereference(br_should_route_hook);
 301                 if (rhook) {
 302                         if ((*rhook)(skb)) {
 303                                 *pskb = skb;
 304                                 return RX_HANDLER_PASS;
 305                         }
 306                         dest = eth_hdr(skb)->h_dest;
 307                 }
 308                 /* fall through */
 309         case BR_STATE_LEARNING:
 310                 if (ether_addr_equal(p->br->dev->dev_addr, dest))
 311                         skb->pkt_type = PACKET_HOST;
 312
 313                 NF_HOOK(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING,
 314                         dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 315                         br_handle_frame_finish);
 316                 break;
 317         default:
 318 drop:
 319                 kfree_skb(skb);
 320         }
 321         return RX_HANDLER_CONSUMED;
 322 }